Equivariant Hypergraph Diffusion for Crystal Structure Prediction

Crystal Structure Prediction (CSP) remains a fundamental challenge with significant implications for the development of new materials and the advancement of various scientific disciplines. Recent developments have shown that generative models, particularly diffusion models, hold great promise for CSP. However, traditional graph-based representations, where atomic bonds are modeled as pairwise graph edges, fail to fully capture the intricate high-order interactions essential for accurately representing crystal structures. In this work, we propose a novel approach that utilizes hypergraphs to represent crystal structures, providing a more expressive abstraction for modeling multi-way atomic interactions. By adopting hypergraphs, we can effectively capture complex high-order relationships and symmetries, such as permutation and periodic translation invariance, which are crucial for characterizing crystal structures. In this work, we propose the \textbf{E}quivariant \textbf{H}ypergraph \textbf{Diff}usion Model (\textbf{EH-Diff}), a generative model designed to take advantage of the symmetry-preserving properties of hypergraphs. EH-Diff exploits these features to offer an efficient and accurate method for predicting crystal structures with a strong theoretical justification to preserve invariance properties. Empirically, we conduct extensive experiments on four benchmark datasets, and the results demonstrate that EH-Diff outperforms state-of-the-art CSP methods with only one sample.
Comment: 14 pages, 4 figures